How to do curve fitting with Weibull, Pseudo logistic and Sigmoidal functions in Excel ?

Copper Contributor

Data of a ParticipantData of a Participant

 

Hi Everyone,

I want to ask how i can do curve fitting of this scatter plot (both graphs). I tried using the option of Trendline but it is showing very few options like exponential, linear, polynomial, logarithmic, power etc. I don't need to use these curve fitting function, rather I need to do curve fitting with Weibull, Pseudo logistic and Sigmoidal functions with max R square value. Please help me out with this how i can implement these mentioned functions in excel. 

Thanks in advance !!!

6 Replies

Pretty sure I can help out if I know to what your objective is. E.g., are you planning to generate random variates from the fitted distributions? IF you simply want to interpolate, you could always do a cubic spline fit, for which I can post code (not sure if I can do that in this forum). Don't use code I posted years back!

Why not use the moments of your data to derive your parameters if you want to stick with parametric models?

 


My objective is very simple. I used the scatter plot for my data set. Got the S shaped curve (shown in figure above). I just want to do curve fitting with these data sets using Weibull or Pseudologistic or sigmoidal function. I am not from mathematics background so i don't understand your question. Sorry for that

@Ankit27 

Ok, it seems you want econometric models, for which you will probably need to use maximum likelihood to calculate the parameters of the distributions. Not that it can't be done in Excel, but I don't remember how to go about fitting the logistic function to the data (and the logistic IS sigmoidal, so I don't understand part of your question). But this isn't an Excel issue, but one of econometrics (or biostatistics, ...). You'll have better luck getting a book on the subject and finding a user group on, say, usenet, replete with statisticians.

I asked for your objective in order to get an idea of the intended use for your curve-fits. There are so many criteria that determine what you even mean by a fit. Some objectives - say Monte Carlo simulation - even obviate the need for a fit.

When posting, explain the axes; are you looking for, say,how  the probability of the cream being useful varies with time of application? That points towards the logistic (aka logit) model; I don't see how the Weibul comes into the picture, which isn't to say it doesn't.

Excel's trendline/regression routines are using ordinary least squares for the fits. You cannot adapt them to what you stated you wanted: again, someone who knows statistics well will almost certainly recommend maximizing the likelihood (in this case the log-likelihood) of your data, which by the way seem overly aggregated for the task.

There are other ways to "fit a curve" that are nonparametric and might be more useful. Posting your intended use of the curve will make answering your question easier.

Groups dealing with R (the language) might be the first place to look. Given the breadth of topics on Usenet, there are probably groups just devoted to statistics. Worth a look. You could also try a local university: a lot of marketing professors and economists are well-versed in the use of the logit model, among others. And perhaps try Redit?

Hope this helped.

There isn't (AFAIK) a pseudo logistic curve. You might be referring to a logistic curve-fit that includes a pseudo R-square. The fit is still maximum-likelihood.